Bachelor's and Master's Theses or Research Internship / Bachelor- und Masterarbeiten oder Forschungspraxis

Benchmarking State-of-the-Art, Graph-based Machine Learning Solvers for Distribution Grid Power Flow

This topic uses a data-driven ML framework developed at TUM EMT to assess the generalization performance and tradeoffs of Graph Neural Network based AC Power Flow solvers in distribution grids. Generalization refers to the solver’s ability to maintain stable and accurate performance when applied in new contexts, i.e. unseen distribution grids. The student will re-implement advanced power flow solvers from literature (including attention- and physics-based GNNs) and perform a large-scale evaluation of the models, including their robustness.

Research method: Prototyping

Research question:

• Can we quantify the generalization potential of the state of the art GNN power flow solvers to determine which are the most robust for grid learning?

• What are the tradeoffs between model choices? Ex. Generalizability vs computational speed vs supervised/unsupervised.

• Can we establish a standard for future distribution grid solver techniques?

Possible approach:

• Literature review on state of the art GNN-based power flow.

• Adoption of open-source models or re-implementation of close-sourced models
• Evaluation and model generalization quantification using provided framework
• Evaluation of other model tradeoffs (i.e. computation, learning assumptions, sensitivity)
• Interpretation of practical implications of the results

Ehimare Okoyomon
e.okoyomon@tum.de

EnergyOS: Operating System Internal Resource and Data Management

Research method: Literature Review, Prototyping

Research question:

How can we store and retrieve data most efficiently?
Can we integrate caching and parallelization of requests?

Possible approach:

• Literature review on data access and storage for time series applications, and data caching.

• Creation and comparison of data management schemes
  • e.g. resource data stored using: (1) python data structures (local dictionaries and lists), (2) json files, (3) csv files, (4) common database like Postgres or MongoDB, (5) specialized time series database like InfluxDB.

• Evaluation of schemes, their performance, how they scale (i.e. complexity in terms of time and storage).

• Design of data caching procedure and parallelization of OS tasks.
• Implementation of Resource and Data Management design in EnergyOS.

Your background/interests:

• Energy Management Systems
• Data Management / Databases
• Programming experience helpful (python)

Ehimare Okoyomon
e.okoyomon@tum.de

EnergyOS: Scheduling and OS-controlled Schedule Generation in a Multi-App EMS Platform

Research method: Literature Review, Prototyping

Research question:

What level of control should the EnergyOS have in creating and executing operation plan schedules for energy components?
How can this be used in practice and what are the advantages and challenges of such an approach?

• Literature review of common EMS scheduling / optimization problems and state of the art methods for solving.

• Implementation of one or more scheduling solutions in EnergyOS.
• Creation of an interface to provide EnergyOS with a set of constraints, objectives, and data/resources.
• Selection and implementation of use cases (EnergyOS apps) to demonstrate how OS-controlled scheduling works.

Your background/interests:

• Energy Management Systems
• Optimizations
• Programming experience helpful (python)

Ehimare Okoyomon
e.okoyomon@tum.de

Development of Smart Grid Simulator (Project EDGE)

Research method: Prototyping

Research questions:

• How can detailed simulation models of energy system components (flexible loads, batteries, smart meters, distribution grid components) be containerized and deployed on a compute cluster?
• How can performance, scalability, and reliability of this simulation infrastructure be measured?
• How can basic EMS scenarios be simulated and analyzed using this infrastructure? 

Possible approach:

• Setup local container development and cluster environment (Docker, Kubernetes, etc.)

• Develop several energy system components as Docker containers exposing web-based interfaces
• Implement message-based communication between containers running on cluster
• Measure performance in different scenarios  

Resources:

• Docker
• Kubernetes
• lemlab
• RabbitMQ

Christoph Goebel

christoph.goebel@tum.de

Sebastian Eichhorn

sebastian.eichhorn@tum.de

Comparing Data Valuation Methods

Research method: prototyping/ simulation

Research question:

• What is the value of a datapoint? How much does each datapoint contribute to model training?
• How do different data valuation measures proposed in the literature compare to each other?

Possible approach: 

• Understand state of the art based on literature
• Implement methods in python
• Compare on various datasets

Resources:

• What is your data worth? - Data Shapley Video

• Data Valuation without training of a model - Research Paper

Jan Marco Ruiz de Vargas

janmarco.ruiz@tum.de 

Extending Data Valuation to Regression Tasks

Research method: prototyping/ simulation

Research question:

• What is the value of a datapoint? How much does each datapoint contribute to model training?
• How can we extend data valuation measures proposed in the literature from classification tasks to regression tasks?

Possible approach: 

• Understand state of the art based on literature
• Think about how to extend data valuation to regression
• Implement methods in python and compare on various datasets

Resources:

• What is your data worth? - Data Shapley Video

• Data Valuation without training of a model - Research Paper

Jan Marco Ruiz de Vargas

janmarco.ruiz@tum.de 

Hypergraph-Guided Decomposition for Multi-Scale Unit Commitment Optimization

Research method: prototyping/ simulation

Research question:

• How can hypergraph inference method be leveraged to inform problem decomposition for multi-level unit commitment and economic dispatch optimization?

• Does utilizing a data-driven optimization problem hypergraph representation could improve accuracy and cut iterations compared with standard partition?

Possible approach: 

• Understand state-of-the-art models based on literature
• Reconstruct optimization hypergraph representation based on time-series solution of the UC+ED optimization problem using SOTA method
• Formulate decomposition method based on the reconstructed hypergraph to form the master and sub-problem partition
• Develop Julia code using state-of-the-art method to solve the problem while utilizing the developed decomposition results
• Evaluate performance and benchmark against SOTA

Resources:

[1] Delabays, Robin, et al. "Hypergraph reconstruction from dynamics." Nature Communications 16.1 (2025): 2691.

[2] Cole, D. L., Gangammanavar, H., & Zavala, V. M. (2023). Hierarchical Graph Modeling for Multi-Scale Optimization of Power Systems. arXiv preprint arXiv:2309.10568.

Arbel Yaniv

arbel.yaniv@tum.de